Elium

What is Elium?

Elium is an enterprise knowledge management platform operated by the Belgian company Whatever SA; it serves as a governed knowledge layer for RAG and AI agents. It is used to organize knowledge scattered across documents, chats, cloud storage, and employee experience into a verified, continuously maintained, and access-controlled repository of content.

The target users of this product and service are teams responsible for global operations, customer support, IT support, AI and data handling, as well as strategic research. The focus is not on individual notes, but rather on ownership, validity period, approval processes, retrieval, distribution, and auditing of corporate knowledge.

Main functions

Knowledge acquisition and organization

Elium refers to the rich-text content created on its platform as articles; these articles can contain text, files, links, videos, and cloud documents. Once published, such content is indexed and can be organized by location, tags, owner, and permissions.

• Use space to divide departments, regions, products, or knowledge areas.
• Use tags to establish thematic, procedural, and audience categorizations across spaces.
• Standardize the formats for SOPs, fault handling, policies, and project retrospectives through templates.
• Retain drafts, versions, annotations, comments, and a history of content changes.
• Identify subject experts and content owners through the personnel directory.
• Use open formats such as Markdown to import or export content.

The structure of the knowledge base should be designed around the user’s tasks, rather than simply copying the existing folder hierarchy. After import, it is still necessary to remove duplicates, assign owners, and set up review periods.

Knowledge lifecycle and governance

For each piece of key knowledge, it is possible to assign a responsible person, set an expiration date, and define verification and approval procedures; expired content will receive alerts or be placed in a processing queue. Corporate and Enterprise versions offer additional features such as multi-step approval, content requests, gap detection, duplicate detection, and automatic transfer.

1. Identify the scope of knowledge that needs to be managed, such as policies, product descriptions, ticket responses, or operation manuals.
2. Create spaces, tags, permissions, and templates to define who can read, edit, and approve.
3. Import existing content and identify duplicate, missing, invalid, and unassigned entries.
4. Assign owners, reviewers, expiration dates, and approval procedures for high-risk knowledge.
5. After publication, data on searches, feedback, and AI responses is collected to identify content that users are unable to find.
6. Forward the gaps and incorrect routing to the responsible person to carry out updates and re-approval.
7. Archive expired versions and retain necessary audit and change records.

Automatic detection can only indicate possible gaps, duplicates, or deviations; the final decision still needs to be made by human personnel. For security, legal, medical, or financial processes, clear mechanisms for manual approval and emergency reversal must be in place.

AI search and answer with citations

Users can ask questions directly, and Elium will perform semantic search among the articles, attachments, and video content to which it has access, in order to generate answers that include citations. The help documentation states that the Answer feature selects the paragraphs with the most similar semantics as the context for the answer.

• The answer uses only information to which the questioner already has access.
• The answer includes citations, allowing users to return to the original text to check the context.
• The search range can be limited to a specified space or knowledge set.
• The feedback mechanism helps knowledge teams identify incorrect answers and missing content.
• Expired or unverified source texts can still lead to seemingly reasonable incorrect answers.
• \"Including citations\" does not mean that the conclusions are necessarily accurate; it is important to check the original text for accuracy.

The value of Elium lies in using well-managed corporate knowledge to constrain the responses provided, rather than ensuring that the model never makes mistakes. Organizations should continuously monitor topics that lack answers, have low confidence levels, involve negative feedback, or pose high risks.

AI editing and automatic translation

AI Editor can expand drafts, improve the way text is expressed, correct spelling and grammar errors, and translate content; it helps non-professional writers create knowledge using standardized templates. Automatic translation enables teams in different locations to access the same knowledge, but technical terms and legal expressions still require review by native speakers.

• First, provide clear facts, terminology, the target audience, and the desired format.
• Limit AI modifications to expression and structure, and prevent it from drafting unverified policies.
• Check each item such as numbers, dates, responsible persons, and operating conditions one by one.
• Create a glossary and have the regional managers review the key language versions.
• Manual approval and version tracking are retained to facilitate the identification of the causes of errors.

Smart Assistants

Smart Assistants are enterprise assistants that are configured using specific prompts and defined knowledge ranges; they can generate responses and drafts for customer service, IT, sales, or research tasks. Each assistant should have access only to the information necessary to complete its tasks.

Custom assistants are part of the Enterprise features; the specific scope of their use is determined by the contract and platform settings. Assistants should not be authorized to carry out actions that exceed their knowledge limits or that require human confirmation due to their high impact.

MCP, API, and enterprise RAG

Elium can provide knowledge to external agents or enterprise RAG systems through three methods: MCP, semantic search, and direct queries to Elium AI. MCP utilizes OAuth 2.0 and carries out queries based on the permissions of the currently logged-in user.

MCP access process

1. The administrator creates connected applications within the Elium tenant.
2. Register the callback address and client information for trusted external assistants.
3. Add the tenant’s MCP service to the compatible client.
4. Use OAuth for login authorization; personal passwords are not shared.
5. First, use a test account with limited permissions to verify the range of content that can be retrieved.
6. It tests scenarios such as authorized tokens, tokens without permissions, expired tokens, and revoked access.
7. Audit external assistant queries, response citations, and abnormal access.

MCP is an open protocol, but Elium’s hosted MCP service does not become open-source as a result of this. Dynamic client registration needs to be enabled on demand, and the callback addresses must be included in an approved list.

Browser extensions and ticketing tools

Corporate and Enterprise versions offer Chrome and Edge extensions that allow knowledge searches, decision trees, and Smart Assistants to be displayed in the sidebar of web pages. Customer service agents can view relevant information alongside tools such as Zendesk, Intercom, EasyVista, JIRA, or GLPI, and use it to draft responses.

• You can view promotional content, save articles, and access the full knowledge base.
• The search results, like those on the main platform, are governed by account permissions.
• When no answer is found, a request can be submitted to add the content to the editing queue.
• The responses to tickets generated by the assistant can be edited before being copied and sent.
• Extensions handle identifiable personal information, and administrators should evaluate browser permissions.
• Managed devices should have their extensions installed, updated, and removed via enterprise policies.

Platforms and applications

Prices and packages

All plans include unlimited space, content, and AI-powered Q&A; no credits are used based on AI usage. Charges are applied per active user, with a minimum of 10 users. Actual taxes, implementation services, connectors, and customization work shall be subject to the quotes, orders, and settlement pages.

To try it out, it is first necessary to communicate with the team in order to gain access to the actual platform; a credit card is not required. The public website does not specify a fixed number of days for the trial period. Upgrades, downgrades, or cancellation can be requested at any time without any fee, and these changes usually take effect starting from the next billing cycle. The rules regarding refunds have not yet been made public.

Support and SLA

The authorities also stated that the system’s availability over the past five years has been above 99.99%; this represents historical performance and does not equate to the monthly 99.5% availability guarantee specified in Enterprise contracts. When making a purchase, it is necessary to clarify the criteria for measuring availability, maintenance windows, compensation measures, recovery targets, as well as the time zones for support.

Privacy, Security, and Deployment

In SaaS platforms, Elium is used to help subscription-based companies manage employee and content data, with the corporate clients acting as the data controllers. Such data may include names, login information, phone numbers, documents, images, calendars, tasks, events, and knowledge content.

• By default, SaaS is hosted in the European Union, with cloud computing and storage services based in the Belgian region being used primarily.
• You can choose from France’s SecNumCloud infrastructure, a private cloud, or on-premises deployment.
• It is currently certified to ISO/IEC 27001:2022, and offers DPA and security materials.
• Transmission is carried out using TLS 1.2 or 1.3, while cloud storage employs AES for disk-level static encryption.
• It offers SSO, 2FA, SCIM, role-based permissions, audit trails, and malware protection.
• Production access requires strict authorization, and SSO, MFA, immediate approval, and auditing are employed.
• Upon the termination of the contract, it is guaranteed that the individual data of platform users will be deleted within 30 calendar days.
• CRM data for potential customers can be stored for 5 years starting from their last activity.

The embedded external web pages, videos, or cloud documents are hosted by third parties, and their data processing is not fully covered by Elium’s privacy policy. Companies need to examine the connectors, LLM providers, and other sub-processors as well.

Differences in terms regarding the use of AI data

The latest version of the Trust Center states that customer content will not be used to train or fine-tune the Elium models or those of its suppliers; it also says that contractual provisions ensure zero data retention for generative AI requests made to each LLM supplier. However, the terms of use for AI specify that the collected data can be used to train and improve the LLMs within the AI functions, with the provision that it is intended solely for use by the respective subscriber and not shared with other customers.

The same AI terms state that the data transmitted to Azure OpenAI is not used to improve OpenAI models, Microsoft products, or general Azure models; the customized models intended for specific customers are available only for use by those subscribers. The scope of these provisions is not entirely consistent.

• The order or AI appendix should specify whether training, fine-tuning, or only retrieval enhancement is to be carried out.
• It is specified which parts of prompts, context, outputs, vectors, and diagnostic logs are subject to zero data retention.
• Confirm the Azure region, model deployment, retention exceptions, and deletion deadline.
• Verify the logs generated when the built-in agent is accessed via MCP or API, as well as the processing carried out by third parties.
• The signed DPA, order, AI terms, and written configurations take precedence over marketing descriptions.

Content rights and restrictions on AI use

Within the limits permitted by applicable law, the AI terms acknowledge that subscribers have the right to provide input, and they grant subscribers rights related to the output generated. This output can be used for legitimate commercial and publishing purposes, but the subscriber is responsible for ensuring compliance and addressing any risks related to third-party rights.

• Before releasing it to the public, it is necessary to indicate that the content was generated by an AI tool.
• The output may be inaccurate, and it might be the same as or similar to the results of other users.
• It is prohibited to mislead others into believing that AI-generated content was created entirely by humans.
• Uses such as causing high-risk economic damage, providing personalized financial or health advice without professional review, are prohibited.
• Privacy, copyrights, trademarks, trade secrets, or other third-party rights must not be violated.
• It is not possible to reverse-engineer the model or extract AI functionality data through unauthorized scraping methods.

API, GitHub, and open-source status

Suitable for users and scenarios

• Global Operations Team: Standardizes SOPs, specifications, handovers, and multilingual processes across all locations.
• Customer Support Team: Generates responses backed by approved policies and FAQs.
• IT team: Develops operation manuals, configurations, solutions for faults, and post-incident reviews.
• AI and Data Team: Provides a governed knowledge layer for RAG, Copilot, and agents.
• Strategy and Research Team: Maintains market intelligence, analysis reports, and decision-making materials.
• Public institutions and regulated enterprises: Use EU sovereign clouds, private clouds, or on-premises deployments.

Advantages and main limitations

Main advantages

• Bring content creation, lifecycle management, AI-based Q&A, and agent integration together on the same platform.
• Permission-aware retrieval and referencing help to reduce unauthorized responses and untraceable answers.
• It supports MCP, GraphQL, browser extensions, mobile apps, and enterprise connectors.
• Options include EU hosting, ISO 27001, SecNumCloud, and on-premises deployment.
• Charging is per user, with unlimited AI Q&A, making it easy to estimate the basic cost of use.

Main limitations

• A minimum of 10 users is required; the initial cost is high for individuals or small teams.
• The implementation period can range from 8 to 16 weeks, and complex projects require efforts in content governance and training.
• Complete APIs, custom assistants, local deployment, and SLAs are available in Enterprise.
• AI-based Q&A systems rely on the quality of the underlying knowledge, and they are unable to eliminate errors caused by outdated information.
• The terms related to training and retention in the Trust Center and the AI Terms of Use do not exactly match.
• Refunds, the fixed trial period, and the costs of certain professional services have not yet been made public.

Frequently Asked Questions

Is Elium suitable for personal knowledge management?

It is primarily aimed at corporate teams, requiring a minimum of 10 users, and includes services for knowledge management, permission control, and implementation. It is not designed mainly for personal note-taking purposes.

Will Elium’s AI responses indicate the source of information?

Yes, the AI search and Answer functions provide citations and allow users to return to the original content for verification. Important conclusions still require checking to ensure their validity and approval.

What is the difference between Team and Corporate?

Corporate includes additional approval processes, checks for gaps and duplicate tests, external portals, browser extensions, integration with proprietary AI systems, SSO, as well as audit trails. Team offers core knowledge and AI capabilities.

How much is Elium?

For annual billing, the cost is 15 euros per user per month for Team plans, and 20 euros for Corporate plans; for monthly billing, the costs are 18 euros and 24 euros respectively, with a minimum of 10 users. Enterprise plans come with customized pricing.

Is there a free version?

There is no confirmed permanently free team version. The platform offers a trial period without the need for a credit card, but it is necessary to discuss this with the team first; the duration of this trial period is not disclosed.

Is there a limit on the number of times AI can be used for answering questions?

The current price page indicates that for all plans, there is no limit on the number of AI queries that can be made, and no usage points are required. Fair use policies, performance requirements, and contractual restrictions still need to be confirmed in the order.

Can it be connected to ChatGPT or Claude?

Compatible assistants can be connected via MCP, and Elium knowledge can be retrieved using OAuth and the permissions of the current user. The permissions and data policies of the external assistants themselves also need to be reviewed.

Does Elium support local deployment?

Supported. Enterprise can choose a private cloud, a French sovereign cloud, or on-premises deployment; the specific architecture, maintenance requirements, and pricing need to be determined on a case-by-case basis.

Will the customer’s content be used to train the model?

The latest Trust Center states that this technology will not be used for training, and no data is retained as a result of such calls; however, the AI-related terms still include provisions concerning customer-specific training. Companies should specify the actual scope in the documents they sign.

Is Elium open source?

The core platform is not a product with publicly available source code. MCP is an open standard, but this does not mean that the Elium servers, search engine, or AI functions are open source.